Fig. 2

Empirical benchmarking of allele-specific read alignment reveals reduced reference bias. a Graphical representation of MEA’s unified strategy for detecting allele-specific reads from RNA-, ChIP-seq and WGBS datasets. Aligning F1 hybrid reads to a pseudogenome enables alignment to their cognate genome even when originating from highly variable loci. b Paired-end WGBS reads (101 bp) from a previously published dataset of C57BL/6 J x DBA/2 J ICM cells [11] were aligned using the Bismark aligner to the (haploid) reference genome (mm10 build) and a MEA-constructed diploid pseudogenome. When using MEA, multiple (2 or more) alignments reflect non-allelic reads, while uniquely aligned reads are allele-specific. Reads aligning uniquely to the pseudogenome were extracted and retroactively assigned to their parental haplotype. c The percentages of allele-specific reads called for each parental haplotype and the number of aligned reads that did not overlap with a genetic variant (non-allelic) is shown. d Allelic contribution of read alignments to each parental haplotype (C57BL/6 J or DBA/2 J) on each autosome. Relative to the script employed by Wang et al. [11], MEA displays about half the reference bias on the majority of autosomes. e Global reference bias for each pipeline is shown